AI for Insurance Underwriting in the Real World

Even as a “maturing” tech, AI still has its critics. And, largely thanks to the sensational public launch of ChatGPT in late 2022 and the proliferation of popular public Large Language Models (LLMs), some of them have a point.  

“Hallucinations” happen and even the latest versions operate in ways that can be opaque to human overseers. Not ideal in a business like insurance, which demands precision. However, a properly trained AI is an ideal solution for solving a long-standing problem for underwriters and others in the industry: data overload.

Underwriting has always been a data business. The more information you have about a risk, and the better your ability to process it, the more accurately you can price it, and the faster you can deliver quotes to win business.  

And while some industry standard solutions have emerged to help increase the pace of document ingestion (e.g., ACORD forms), there are countless non-standardized forms to deal with, plus unstructured documents that might include loss runs, SOVs, equipment schedules and even geospatial information or satellite images. At present, weaving all that unstructured data into workflows is time-consuming and inexact; or in some cases doesn’t happen at all.

One unfortunate outcome from all this data needing manual processing is that underwriting professionals spend just 30% of their time doing risk analysis, according to a landmark survey produced by Accenture. Underwriters are spending their valuable time reviewing submission documents looking for “important data siloed in PDFs and spreadsheets attached to emails from brokers.”

Among the more concerning conclusions in this study: “To assess risk, underwriters still have to move between different documents, looking for data that’s formatted in different ways depending on the broker it’s coming from.” It’s tedious, repetitive work. Worse still, it diverts underwriters' focus aways from value-enhancing work, such as customer relationship-building and creating innovative products.  

“Underwriters are fed up”

As a result, underwriters are fed up. When polled by The Institutes and Accenture, just 26% of them gave a positive rating to their firm’s talent management performance  – down from 56% in 2013.

Enter artificial intelligence – or, more specifically, Generative AI systems that have been expertly trained around insurance and purpose-built for underwriting.  

Insurance-specific data and learning is critical for an AI solution to deliver real value. As one engineer at OpenAI (whose ChatGPT product helps job-seekers craft cover letters   –   and lets kids write thank-you notes – pointed out), “model behavior is not determined by architecture, hyperparameters, or optimizer choices. It’s determined by your dataset, nothing else.”

This explains why insurers can’t just point an off-the-shelf LLM like ChatGPT at underwriting workflows and expect it to deliver results. Generic LLMs look at universal datasets to create convincing, but merely probabilistic, strings of text, pictures and even music (hence, hallucinations). Targeted LLMs such as Roots Automation’s InsurGPT™ are trained on the largest set of insurance-specific data, to “understand” unstructured information, including photographs, emails and social media posts, to arrive at accurate conclusions about how to handle it.

A problem as old as insurance: too much data, too little time

The problem is clear: too much data, too little time. A likely solution is at-hand: Generative AI that can formalize unstructured data. However, implementing tools built on the technology still looms as a huge challenge for individual organizations.

A recent Oliver Wyman/Celent poll revealed roughly half of all insurance firms are experimenting with Generative AI already (with 86% of the biggest firms dipping a toe in the water). Nevertheless, a tiny fraction of them – approximately 9% – have implemented it in a “product environment.”

The missing link between AI on the workbench and AI in the workflow is insurers' ability to work with people who understand both worlds – insurance and AI – and can harness the new tech to deliver fast, accurate and reliable performance.  

That way, underwriters aren’t faced with costly and uncertain in-house AI development projects. And they also don’t get left behind as the industry evolves new ways to accelerate underwriting data solutions.

That’s the big idea behind our AI-Powered Commercial Underwriting Digital Coworker.  

By understanding the data and workflow problems facing insurance organizations and then training our systems on millions of documents containing highly contextualized insurance data, we’ve built an AI solution to solve complex data processing workload challenges.  

‍

Roots customers that have implemented Digital Coworkers have realized significant improvements across their underwriting practices, such as 25% reduction in underwriting workload and 10% reduction in underwriting cost. Want to know more about how to we’re helping return underwriters to high-value tasks by eliminating up to 80% of the work from document processing and data management?

Download your copy of our latest white paper, Transforming Commercial Underwriting with Digital Coworkers for insights into obtaining and deploying the tools and technologies that are driving today’s insurance technology revolution.